Various array normalization methods have been developed for gene expression microarrays. Most of these methods assume few or symmetric differential expression between sample groups. There has been no systematic study of the properties of these methods in normalizing microRNA expression arrays utilizing heterogeneous samples such as tumors. MicroRNA arrays contain only a few hundred microRNAs, and are likely to have a relatively large proportion being differentially expressed between diverse tumor groups. The assessment of normalization methods in this setting is difficult because of the lack of a benchmark dataset that has no confounding array effects. We propose to design and generate such benchmark datasets, perform a systematic assessment of normalization methods with a particular emphasis on the utility of these models for detecting markers with differential expression, and from the benchmark data design derive statistical models that acknowledge heterogeneities inherent to tumor samples. PUBLIC HEALTH RELEVANCE: Microarrays are being widely used in cancer research. A critical step for processing microarray data is to normalize the arrays so that measurements from different arrays are comparable. There is a great need to evaluate the properties of statistical methods for array normalization when they are applied to microRNA arrays utilizing heterogeneous samples such as tumors.